The point on the OpenAI research acceleration report is particularly apt. Like it shows time horizons for 80% accuracy of ~15 minutes, that's way too low for such short timelines to make sense.
I think waking people up to urgently working on this now rather than later is robustly good. it’s not as if we have a finite pool of resources allocating to short and long timelines. The short timeline people are most likely growing that pool the most.
I kind of agree, but not entirely. It's true that getting more resources and people working on this is, on the net, good. But de-prioritizing longer events is a real risk.
As a toy example: we already see people who have dismissed AI safety because of the (in retrospect quite silly) push from a minority at OpenAI to not publish GPT-2. Focusing too much on 27 could backfire in the same way but at a much larger scale. If the public spends a year hearing about how superintelligence WILL come by 27, and then 28 rolls around with business mostly as usual, it will be harder to garner support for 29.
We could also face a “Great Disappointment” of our own. I've met a few people who are currently, quite literally dropping everything to work on advocacy. I respect this a lot. But if we became convinced that 27 is THE year ASI comes, then almost everyone should drop everything and work on advocacy. That would clearly have negative consequences on even slightly longer timelines.
The point on the OpenAI research acceleration report is particularly apt. Like it shows time horizons for 80% accuracy of ~15 minutes, that's way too low for such short timelines to make sense.
Re: why this matters.
I think waking people up to urgently working on this now rather than later is robustly good. it’s not as if we have a finite pool of resources allocating to short and long timelines. The short timeline people are most likely growing that pool the most.
I kind of agree, but not entirely. It's true that getting more resources and people working on this is, on the net, good. But de-prioritizing longer events is a real risk.
As a toy example: we already see people who have dismissed AI safety because of the (in retrospect quite silly) push from a minority at OpenAI to not publish GPT-2. Focusing too much on 27 could backfire in the same way but at a much larger scale. If the public spends a year hearing about how superintelligence WILL come by 27, and then 28 rolls around with business mostly as usual, it will be harder to garner support for 29.
We could also face a “Great Disappointment” of our own. I've met a few people who are currently, quite literally dropping everything to work on advocacy. I respect this a lot. But if we became convinced that 27 is THE year ASI comes, then almost everyone should drop everything and work on advocacy. That would clearly have negative consequences on even slightly longer timelines.